CN109606263A - To the method and the blind spot monitoring device for using this method for monitoring that the blind spot of vehicle is monitored - Google Patents
To the method and the blind spot monitoring device for using this method for monitoring that the blind spot of vehicle is monitored Download PDFInfo
- Publication number
- CN109606263A CN109606263A CN201811167644.4A CN201811167644A CN109606263A CN 109606263 A CN109606263 A CN 109606263A CN 201811167644 A CN201811167644 A CN 201811167644A CN 109606263 A CN109606263 A CN 109606263A
- Authority
- CN
- China
- Prior art keywords
- suggestion box
- blind spot
- vehicle
- box
- test
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 74
- 238000012806 monitoring device Methods 0.000 title claims abstract description 51
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000013507 mapping Methods 0.000 claims abstract description 38
- 238000012360 testing method Methods 0.000 claims description 59
- 238000012549 training Methods 0.000 claims description 34
- 238000012545 processing Methods 0.000 claims description 32
- 238000005070 sampling Methods 0.000 claims description 12
- 230000008859 change Effects 0.000 claims description 8
- 238000004891 communication Methods 0.000 claims description 8
- 230000000007 visual effect Effects 0.000 description 15
- 238000001514 detection method Methods 0.000 description 12
- 238000013527 convolutional neural network Methods 0.000 description 7
- 238000012790 confirmation Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 210000000056 organ Anatomy 0.000 description 3
- 238000011897 real-time detection Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000000151 deposition Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 238000001035 drying Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 229910044991 metal oxide Inorganic materials 0.000 description 1
- 150000004706 metal oxides Chemical class 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/223—Analysis of motion using block-matching
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60R—VEHICLES, VEHICLE FITTINGS, OR VEHICLE PARTS, NOT OTHERWISE PROVIDED FOR
- B60R1/00—Optical viewing arrangements; Real-time viewing arrangements for drivers or passengers using optical image capturing systems, e.g. cameras or video systems specially adapted for use in or on vehicles
- B60R1/02—Rear-view mirror arrangements
- B60R1/025—Rear-view mirror arrangements comprising special mechanical means for correcting the field of view in relation to particular driving conditions, e.g. change of lane; scanning mirrors
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
- B60W30/18—Propelling the vehicle
- B60W30/18009—Propelling the vehicle related to particular drive situations
- B60W30/18163—Lane change; Overtaking manoeuvres
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W40/00—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models
- B60W40/02—Estimation or calculation of non-directly measurable driving parameters for road vehicle drive control systems not related to the control of a particular sub unit, e.g. by using mathematical models related to ambient conditions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/20—Image enhancement or restoration using local operators
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
- G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters
- G06V10/451—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters with interaction between the filter responses, e.g. cortical complex cells
- G06V10/454—Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60R—VEHICLES, VEHICLE FITTINGS, OR VEHICLE PARTS, NOT OTHERWISE PROVIDED FOR
- B60R2300/00—Details of viewing arrangements using cameras and displays, specially adapted for use in a vehicle
- B60R2300/30—Details of viewing arrangements using cameras and displays, specially adapted for use in a vehicle characterised by the type of image processing
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60R—VEHICLES, VEHICLE FITTINGS, OR VEHICLE PARTS, NOT OTHERWISE PROVIDED FOR
- B60R2300/00—Details of viewing arrangements using cameras and displays, specially adapted for use in a vehicle
- B60R2300/80—Details of viewing arrangements using cameras and displays, specially adapted for use in a vehicle characterised by the intended use of the viewing arrangement
- B60R2300/802—Details of viewing arrangements using cameras and displays, specially adapted for use in a vehicle characterised by the intended use of the viewing arrangement for monitoring and displaying vehicle exterior blind spot views
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60Y—INDEXING SCHEME RELATING TO ASPECTS CROSS-CUTTING VEHICLE TECHNOLOGY
- B60Y2400/00—Special features of vehicle units
- B60Y2400/30—Sensors
- B60Y2400/301—Sensors for position or displacement
- B60Y2400/3015—Optical cameras
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/12—Bounding box
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Databases & Information Systems (AREA)
- Evolutionary Biology (AREA)
- Mechanical Engineering (AREA)
- Automation & Control Theory (AREA)
- Transportation (AREA)
- Biodiversity & Conservation Biology (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
- Closed-Circuit Television Systems (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of method monitored using blind spot of the blind spot monitoring device to monitoring vehicle.The described method comprises the following steps: blind spot monitoring device (a) obtains Feature Mapping from rear video image under the following conditions: obtaining the video image about the reference vehicle in blind spot, creation refers to the reference block of vehicle, and reference block is set as Suggestion box;(b) feature vector for Suggestion box in Feature Mapping is obtained by pondization, feature vector is input to and is fully connected layer, obtain classification and return information;And Suggestion box (c) is selected by referring to classification information, the bounding box for being used for Suggestion box by using acquisition of information is returned, confirm bounding box Suggestion box matching whether corresponding with they, and determines whether monitored vehicle be located in blind spot in Suggestion box with determining monitored vehicle.
Description
Technical field
The present invention relates to the method and the blind spot monitoring device for using this method for monitoring that the blind spot of vehicle is monitored.More
Body, it is related to a kind of method that the blind spot to monitoring vehicle is monitored and the blind spot monitoring device for using this method, the method
The following steps are included: (a) is under conditions of blind spot monitoring device completes following processing: (i) obtain with range surveillance vehicle unit away from
From and positioned at monitor that the one or more in the blind spot of vehicle refers to the relevant rear video image for sampling of vehicle, (ii) is created
Build with the corresponding reference block of reference vehicle in each of the rear video image for sampling, and (iii) by reference block
It is set as m Suggestion box, the m Suggestion box is used as the candidate region in the blind spot of monitoring vehicle, wherein candidate region has
Detect the probability of at least one monitored vehicle, then if obtaining the rear video image for test from the monitoring vehicle of operation,
Then blind spot monitoring device obtains at least one Feature Mapping for test from the rear video image for test;(b) blind spot is supervised
Visual organ (i) by by pond operation (pooling operation) be applied to for test Feature Mapping on m Suggestion box
Come obtain it is corresponding with each of m Suggestion box for test feature vector each, (ii) will be used to test
Feature vector be input at least one and be fully connected layer, (iii) obtain it is corresponding with each of m Suggestion box each
The classification score for test of class, and (iv) obtain being used for for each class corresponding with each of m Suggestion box
The recurrence information of test;(c) blind spot monitoring device executes following processing: (i) is built by referring to the classification score for test at m
It discusses and selects j Suggestion box in frame, (ii) is believed by using the recurrence for test corresponding with each of j Suggestion box
Breath obtains each of bounding box corresponding with each of j Suggestion box, and each of (iii) confirmation bounding box is
It is no to determine and be monitored equal to or more than first threshold, and (iv) with the matching degree of its corresponding Suggestion box in j Suggestion box
Whether vehicle is located in each of j Suggestion box, thereby determines that whether monitored vehicle is located in the blind spot of monitoring vehicle.
Background technique
It monitors that every side of vehicle has side-view mirror, and there is rearview mirror in the front, center in its cabin, for driving
Side and the subsequent good visual field needed for member changes lane.
Although side-view mirror for observing every side and behind, its with driver cannot see that monitored vehicle or
Blind spot (BS) of any other very close object with it.
This has become a problem, because if driver changes vehicle in the case where not seeing monitored vehicle
Road is monitored vehicle in blind spot, then accident may occur.
Convex mirror is placed on the corner of side-view mirror by such problems in order to prevent, driver sometimes, this to drive
Member is it can be seen that blind spot.
However, even if driver also must see blind spot with the eyes of oneself to change when convex mirror is added on side-view mirror
Become lane, this brings bigger pressure to driver, even and if there may be the head position that driver changes oneself is logical
Cross the blind spot that convex mirror still can not be seen.
In order to prevent this situation, being recently proposed a kind of blind spot monitoring system, it is intended to by using positioned at monitoring vehicle
The sensor at rear portion to driver provide about be located at blind spot or close to blind spot monitored vehicle detection information, to prevent
The only accident when driver does not have to notice the monitored vehicle in blind spot and changes lane.
Particularly, being generallyd use using the blind spot monitoring system of visual sensor being capable of several spy of view-based access control model infomation detection
The algorithm of sign.
However, these algorithms may show the limited inspection limited by external environment, body form and system structure
Survey rate.Since accurate detection needs many visual processes, so computation burden is very heavy.Therefore, because limited processing money
Source, real-time detection may be difficult in embedded systems.
Cause in convolutional neural networks (CNN) slow-footed main problem first is that network (RPN) is suggested in region.In order to
Candidate is extracted from final Feature Mapping, RPN determines whether sliding window includes candidate in each position.It is fully connected
(FC) determine whether candidate is vehicle, however, many candidates overlap each other and RPN consumes a large amount of runing times to these
The redundancy candidate for being almost helpless to improve verification and measurement ratio executes calculating.
As another example for using visual sensor to detect monitored vehicle, exist a kind of by motion vector expression view
Feel the optical flow approach of the movement of pixel.However, identifying that the algorithm of monitored vehicle is largely dependent upon using optical flow approach
The variation of background state and visual noise, and huge calculated load is needed, thus it is not easy to carry out the reality of monitored vehicle
When detect.
Summary of the invention
It is an object of the invention to solve all above problems.
Another object of the present invention is that easily detection is located at the monitored vehicle in blind spot.
A further object of the present invention is monitored vehicle of the detection in blind spot without regard to driving environment.
A further object of the present invention is to provide a kind of algorithm, and the algorithm is for detecting the monitored vehicle being located in blind spot
Need less calculating.
A further object of the present invention is the monitored vehicle that real-time detection is located in blind spot, without regard to background state
With the variation of visual noise.
A further object of the present invention is the monitored vehicle accurately detected in blind spot using CNN.
According to an aspect of the present invention, provide it is a kind of by using blind spot monitoring device to monitoring vehicle blind spot supervise
Depending on method, comprising the following steps: (a) is under conditions of blind spot monitoring device completes or support another device to complete following processing:
(i) it is relevant with reference to vehicle to obtain to range surveillance vehicle unit distance and be located at the one or more in the blind spot of monitoring vehicle
For the rear video image of sampling, (ii) creation is corresponding with the reference vehicle in each of the rear video image for sampling
Reference block, and reference block is set as m Suggestion box by (iii), and the m Suggestion box is used as in the blind spot of monitoring vehicle
Candidate region, wherein candidate region has the probability for detecting at least one monitored vehicle, then if from the monitoring vehicle of operation
The middle rear video image obtained for test, then blind spot monitoring device is obtained or is supported from the rear video image for test and is another
Device obtains at least one Feature Mapping for test;(b) blind spot monitoring device (i) is by the way that pond operation to be used to survey
M Suggestion box in the Feature Mapping of examination obtains and each of m Suggestion box phase to obtain or support another device
The each of the corresponding feature vector for test, (ii) input or support feature vector of another device input for test
Layer is fully connected at least one, it is corresponding with each of m Suggestion box that (iii) obtains or support that another device is obtained
The classification score for test of each class, and (iv) obtain or support another device obtain with it is each in m Suggestion box
The recurrence information for test of the corresponding each class of person;(c) it is following to execute or support that another device executes for blind spot monitoring device
Processing: (i) selects j Suggestion box by referring to the classification score for test in m Suggestion box, (ii) by using with j
The corresponding recurrence acquisition of information for test of each of a Suggestion box is corresponding with each of j Suggestion box
The each of bounding box, (iii) confirm each of bounding box whether with of its corresponding Suggestion box in j Suggestion box
It is equal to or more than first threshold with degree, and (iv) determines whether monitored vehicle is located in any one of j Suggestion box, by
This determines whether monitored vehicle is located in the blind spot of monitoring vehicle.
According to another aspect of the present invention, a kind of blind spot monitoring that the blind spot for monitoring vehicle is monitored is provided
Device, comprising: communication unit, under the following conditions: obtaining with range surveillance vehicle unit distance and in the blind spot of monitoring vehicle
One or more with reference to vehicle it is relevant from monitoring vehicle shoot for sampling rear video image;It creates and is used to sample
Rear video image each in the corresponding reference block of reference vehicle;And reference block is set as m Suggestion box, institute
State the candidate region that m Suggestion box is used as in the blind spot of monitoring vehicle, wherein candidate region has detection, and at least one is monitored
The probability of vehicle, the communication unit is for obtaining or supporting another device to obtain for testing from the monitoring vehicle driven
Rear video image;And processor, for executing or supporting another device to execute following processing: (i) is after for test
The Feature Mapping for test is obtained in video image;(ii) by the Feature Mapping for test with m Suggestion box
The each in corresponding region obtains feature vector corresponding with each of m Suggestion box using pond operation,
The feature vector for being used to test corresponding with each of m Suggestion box is input at least one and is fully connected layer,
(ii-1) the classification score for test of each class corresponding with each of m Suggestion box is obtained, (ii-2) is obtained
The recurrence information for test of each class corresponding with each of m Suggestion box;And (iii) by referring to being used for
The classification score of test selects j Suggestion box in m Suggestion box, by using corresponding with each of j Suggestion box
For test recurrence acquisition of information bounding box corresponding with each of j Suggestion box, confirmation bounding box in it is every
Whether one is supervised with the matching degree of its corresponding Suggestion box in j Suggestion box equal to or more than first threshold, and determination
Whether it is located in any one of j Suggestion box depending on vehicle, thereby determines that whether monitored vehicle is located at the blind spot of monitoring vehicle
In.
In addition, also provide can be readable by a computer for storing computer program to execute remembering for method of the invention
Recording medium.
Detailed description of the invention
For illustrating that the attached drawing below of example embodiments of the present invention is only a part of example embodiments of the present invention,
Those skilled in the art can obtain other figures based on attached drawing without creative efforts.
Fig. 1 is the block diagram for schematically showing blind spot monitoring system according to one example embodiment;
Fig. 2 is the blind spot monitoring system on the monitoring vehicle schematically shown according to one example embodiment
Figure;
Fig. 3 is the block diagram for schematically showing the CNN for executing blind spot monitoring according to one example embodiment;
Fig. 4 is that rear video image of the use for sampling schematically shown according to one example embodiment comes
Figure of the setting for the processing of the Suggestion box of blind spot monitoring;
Fig. 5 is the figure for schematically showing the processing for executing blind spot monitoring according to one example embodiment;
Fig. 6 is to schematically show dynamic Suggestion box of the addition for blind spot monitoring according to one example embodiment
Processing figure.
Specific embodiment
The present invention is carried out in detail below with reference to the exemplary attached drawing of implementable specific embodiments of the present invention and diagram is shown as
Illustrate, so that the object, technical solutions and advantages of the present invention become apparent.These embodiments are described in sufficient detail, so that
Those skilled in the art can implement the present invention.
In addition, term " includes " and its modification are not intended to and exclude other in detailed description of the invention and claim
Technical characteristic, additive, component or step.Other objects of the present invention, benefit and feature will be partially by specifications simultaneously
And it discloses partially by the embodiment of the present invention to those skilled in the art.The following examples and attached drawing will provide conduct
Example, but they are not intended to be limited to the present invention.
In addition, the present invention covers all possible combination for the example embodiment pointed out in this specification.It should be understood that
It is, although various embodiments of the present invention are different, to be not necessarily mutually exclusive.For example, not departing from spirit of the invention
In the case where range, can implement in other embodiments herein in conjunction with one embodiment describe special characteristic, structure or
Characteristic.Also, it should be understood that without departing from the spirit and scope of the present invention, can modify each disclosed
The position of each element in embodiment or arrangement.Therefore, the following detailed description is not be considered in a limiting sense, the present invention
Range be defined solely by the appended claims, and suitably explained together with the full scope of the equivalent of claim.?
In attached drawing, identical appended drawing reference indicates the same or similar function in all figures of the drawings.
In order to make those skilled in the art easily implement the present invention, below in reference to attached drawing to exemplary reality of the invention
Example is applied to be described in detail.
Firstly, Fig. 1 is the block diagram for schematically showing blind spot monitoring system according to one example embodiment.Ginseng
According to Fig. 1, blind spot monitoring system may include blind spot monitoring device (BSM) 100 and monitoring vehicle 200.
Blind spot monitoring device 100 can detecte the blind spot for being located at monitoring vehicle or another vehicle close to blind spot, i.e., monitored
Vehicle.Specifically, blind spot monitoring device 100 can be analyzed by using visual sensor 10 (for example, camera) and be obtained from monitoring vehicle
The rear video image that takes determines whether monitored vehicle is located in blind spot.
Then, it by referring to the information about steering system and the information sent from blind spot monitoring device 100, can prevent
Monitoring vehicle 200 changes lane on the direction for being confirmed as be located at blind spot towards monitored vehicle, or can permit
Warning system warning driver is monitored vehicle and is in the fact in blind spot.It particularly, is autonomous vehicle in monitoring vehicle 200
It, can be by referring to the information about driving environment and about from blind spot monitoring device in the case where (autonomous vehicle)
The information of monitored vehicle in 100 blind spots received come determine whether drive while change lane.
In addition, blind spot monitoring device 100 may include communication unit 110 and processor 120, communication unit 110 uses visual sensing
Device 10 obtains rear video image from monitoring vehicle, and processor 120 judges to be monitored by analyzing acquired rear video image
Whether vehicle is in the blind spot of monitoring vehicle.Here, as shown in Fig. 2, being mounted on the view on any position of monitoring vehicle 200
Feel that sensor 10 can capture the video of its rearview, and may include optical sensor, such as charge-coupled device
(CCD), complementary metal oxide semiconductor (CMOS) or image capture apparatus.
Further, processor 120 can execute following processing: (i) obtains at least one from acquired rear video image
Pond operation is applied to the region corresponding with each of m Suggestion box in Feature Mapping by a Feature Mapping, (ii)
Each, and (iii) obtain each of corresponding with each of m Suggestion box feature vector.Here, m can
To be predetermined value.Then, processor 120 can execute following processing: feature vector is input at least one and is fully connected by (i)
(FC) layer, (ii) obtains the classification score of each class corresponding with each of m Suggestion box, and (iii) is obtained and m
The recurrence information of the corresponding each class of each of a Suggestion box.Later, processor 120 can execute following processing: (i)
By referring to classification score, selection is confirmed as the j Suggestion box with monitored vehicle in m Suggestion box, and (ii) passes through
It is corresponding using corresponding with each of j Suggestion box every recurrence each of acquisition of information and j Suggestion box
Bounding box each, (iii) confirm each of bounding box whether with its corresponding Suggestion box in j Suggestion box
Matching degree is equal to or more than first threshold, and correspondingly (iv) determines whether monitored vehicle is located at appointing in j Suggestion box
In one.Therefore, the information in the available blind spot for whether being located at monitoring vehicle about monitored vehicle.
Below by referring to Fig. 3 to Fig. 6 come more specifically to for monitoring whether monitored vehicle is located at monitoring vehicle
Blind spot in method be illustrated.
Firstly, as shown in figure 3, m Suggestion box is set as candidate region by driver in step S1.Here, candidate region
With the probability for being equal to or more than at least one monitored vehicle of the detection of predetermined value.
As an example, as shown in figure 4, driver it is available about monitoring vehicle blind spot in be located at unit away from
From reference vehicle from monitoring vehicle capture for sampling rear video image.In other words, it is supervised in reference vehicle in distance
In the case where being located at left and right lane at vehicle one or more cell distance, the vision being mounted on monitoring vehicle can be used
Rear video image of the sensor capture for sampling.Here, for convenience, unit distance is set as one meter in Fig. 4, but can also
To be arbitrary value.Unit distance and capture be can change with reference to the quantity of the image of vehicle, so that all possible candidate region
All it is coated in blind spot.
Then, each of reference block P1, P2 and P3 can be created as corresponding respectively to being located at each with reference to vehicle every
Each of rear video image for sampling captured in the case where at a distance is with reference to vehicle, reference block P1, P2 of creation
It can be set as m Suggestion box P1, P2 and P3 as candidate region with P3.In Fig. 3 to Fig. 6, for convenience, the value of m is set
It is 3.
Here, monitoring vehicle left and right side respectively there is blind spot, therefore, by set side m Suggestion box P1,
Other Suggestion box in the blind spot equivalent as the other side of P2 and P3 can set and carry out to the blind spot of monitoring vehicle two sides
Whole group Suggestion box needed for monitoring.
Next, under conditions of setting m Suggestion box P1, P2 and P3 as candidate region as in figure 4, if
The backsight captured from monitoring vehicle is obtained via communication unit 110 when monitoring that vehicle is currently running and (is driven) as shown in Figure 5
Frequency image, then in the step S2 of Fig. 3, blind spot monitoring device 100 can obtain Feature Mapping from rear video image.
As an example, when monitoring vehicle is currently running, if visual sensor 10 captures and sends monitoring vehicle
Rear video image, then the available rear video image sent from visual sensor 10 of communication unit 110, processor 120 can be with
Rear video image is input to convolutional layer, and by obtaining Feature Mapping to rear video image application convolution algorithm.
Here, under conditions of filling (pad) is set as zero, processor 120 can be by the filter slided with predetermined step width
Wave device is to rear video image application convolution algorithm or to the Feature Mapping application convolution algorithm from rear video image creation.This
In, filling is the amount of filter displacement for adjusting the output size from convolutional layer, stride.In other words, processor 120 can be right
Rear video image application convolution algorithm can execute multiple convolution algorithms and be made a reservation for obtaining to have to reduce their size
The Feature Mapping of size.In contrast, under conditions of executing convolution when filling is set in the filling of some nonzero value, place
Reason device 120 can execute following processing: (i) obtains size Feature Mapping identical with the size of rear video image;(ii) pass through
The size of Feature Mapping is reduced using pond operation to acquired Feature Mapping;And (iii) passes through repetition convolution sum pond
Change to create the Feature Mapping with predefined size.
In addition, processor 120 can be in the whole region of each of the rear video image obtained from visual sensor 10
Above or on a part of image corresponding with the region for the blind spot being set as in rear video image volume is executed using convolutional layer
Product, and therefore obtain Feature Mapping.
Next, processor 120 can be in step s3 to opposite with m Suggestion box P1, P2 and P3 in Feature Mapping
Apply pond operation, and available feature vector corresponding with each of m Suggestion box in the region answered.
As an example, processor 120 can execute following processing: (i) by Feature Mapping with m Suggestion box
The corresponding a part of Feature Mapping of each of P1, P2 and P3 is input in the layer of the pond ROI;(ii) Feature Mapping with
M Suggestion box P1, P2 ..., the corresponding each region Pm execute maximum pond or the pond that is averaged;And (iii) is obtained and m
The each of the corresponding feature vector of each of a Suggestion box P1, P2 and P3.
Next, processor 120 can execute following processing: (i) in step S4, will in m Suggestion box P1, P2 and P3
The each of the corresponding feature vector of each be input at least one and be fully connected (FC) layer;(ii) from the first FC layers obtain
Take the classification score of each class corresponding with each of m Suggestion box P1, P2 and P3;And (iii) from the 2nd FC layers
Obtain the recurrence information of each class corresponding with each of m Suggestion box P1, P2 and P3.Here, class can be used for really
The object that regular inspection is surveyed is what or the object for determining detection whether be vehicle.
Next, in step S5, processor 120 can be by referring to the classification score of each class from the first FC layers of acquisition
J Suggestion box in m Suggestion box P1, P2 and P3 is determined as to include monitored vehicle.
Next, processor 120 can be by using corresponding with each of j Suggestion box every in step S6
Item return acquisition of information be determined as including the corresponding bounding box of each of j Suggestion box for being monitored vehicle.
As an example, the processor 120 of blind spot monitoring device 100 can be corresponding with one of j Suggestion box
A part return information in select class corresponding with the class for being appointed as vehicle specific (class-specific) return information,
And specific Suggestion box recurrence can be made to obtain the bounding box for surrounding monitored vehicle by using the specific information that returns of class.
Next, processor 120 can execute following processing: (i) determine bounding box each whether with j Suggestion box
Each of corresponding Suggestion box matching degree be equal to or more than first threshold;(ii) determine whether monitored vehicle is located at j
In at least one Suggestion box in a Suggestion box;And (iii) determines whether monitored vehicle is located in the blind spot of monitoring vehicle.
On the other hand, if being less than second threshold at least with the matching degree of the corresponding Suggestion box in j Suggestion box
One specific border frame is confirmed as being detected, then specific border frame can be set as at least one dynamic by processor 120
Suggestion box, as shown in the DPB (dynamic Suggestion box) in Fig. 5, the dynamic Suggestion box be added to include candidate region group
New element, to obtain update group.Here, first threshold and second threshold can be the same or different from each other.By in addition to using
Also using the dynamic Suggestion box of addition, blind spot monitoring device 100 can by m Suggestion box P1, P2 and P3 on the next frame of subject image
To track specific monitored vehicle.
For example, referring to Fig. 6, if obtaining specific border frame (i.e. DPB) from the t frame of rear video image, processor 120 can
Dynamic Suggestion box and m Suggestion box P1, P2 and P3 to be set as including in the update group on the t+1 frame of rear video image.
Further, if meeting following all conditions: (i) dynamic Suggestion box and m Suggestion box P1, P2 and P3 are set
For candidate region as described above;(ii) vehicle is monitored to be confirmed as being located at the dynamic Suggestion box of (k-1) frame from t+1 frame to t+
In;And (iii) is monitored vehicle and is confirmed as not being located in the dynamic Suggestion box on t+k frame, then processor 120 can pass through
Determine whether monitored vehicle is located in blind spot using m Suggestion box P1, P2 and P3 on t+k+1 frame.
Here, processor 120 can execute following processing: (i) calculates the first overlapping region, the first overlapping region packet
Include the region that the corresponding j Suggestion box of each of bounding box overlaps each other;And (ii) will be equal to or more than third
Some corresponding Suggestion box of specific first overlapping region in first overlapping region of threshold value is determined as including monitored vehicle.
Here, bounding box has dijection corresponding relationship (bijective correspondence) with j Suggestion box.Show as one
Example, if obtaining some bounding box corresponding with specific Suggestion box, processor 120 can calculate some bounding box with
The first Duplication between the specific Suggestion box, that is, pass through the region that (i) some bounding box is Chong Die with the specific Suggestion box
Divided by the ratio that the region of (ii) some bounding box and the sum of the region (union) of the specific Suggestion box obtain, and if
First Duplication is equal to or more than third threshold value, then can be determined as specific Suggestion box including monitored vehicle.
In addition, processor 120 can execute following processing: (i) calculates the second overlapping region that bounding box overlaps each other, often
A bounding box corresponds to each of j Suggestion box;(ii) the second weight that will be confirmed to be be equal to or more than the 4th threshold value
At least one of folded region at least one corresponding specific border frame of specific second overlapping region is determined as including single phase
Same monitored vehicle;And in (iii) maximum specific border frame in region for being overlapped corresponding Suggestion box selected by
The bounding box selected is determined as including monitored vehicle.Here, bounding box has dijection corresponding relationship with j Suggestion box.
Further, processor 120 can execute following processing: (i) calculates the second overlapping region, and (ii) will be identified
For specific border frame corresponding with each of specific second overlapping region in the second overlapping region less than the 5th threshold value
Each be determined as including each monitored vehicle.
As an example, obtain the first bounding box corresponding with the first Suggestion box and the second Suggestion box difference and
In the case where the second boundary frame, processor 120 can calculate the second Duplication between the first bounding box and the second boundary frame,
That is, passing through region and the second side by (i) the first bounding box region Chong Die with the second boundary frame divided by (ii) first bounding box
The ratio that the sum of the region of boundary's frame obtains, and if the second Duplication is equal to or more than the 4th threshold value, processor 120 can be with
First bounding box and the second boundary frame are determined as to include single identical monitored vehicle, and if the second Duplication is less than
5th threshold value, then the first bounding box and the second boundary frame can be determined as including each monitored vehicle by processor 120.This
In, the 4th threshold value and the 5th threshold value can be the same or different from each other.Further, if the first bounding box and the second boundary frame are true
It is set to including single identical monitored vehicle, then processor 120 can be by the between the first bounding box and the first Suggestion box
The 4th Duplication between triple-overlapped rate and the second boundary frame and the second Suggestion box is compared, and if third Duplication is big
In the 4th Duplication, then the first bounding box can be determined as including monitored vehicle.
Next, processor 120 can be by sending control for the information about the monitored vehicle being located in blind spot
Unit, to support the control unit of monitoring vehicle 200 to prevent monitoring vehicle 200 being confirmed as institute position towards monitored vehicle
In blind spot direction on change lane.
Here, want to change lane in driver and monitored vehicle is detected as being located at what driver intended to move to
In the case where blind spot on lane, by referring to the information of the steering system about monitoring vehicle 200, the control of vehicle 200 is monitored
Unit processed can make warning system warning driver be monitored the fact that vehicle is located in blind spot, and driver can be made to avoid
The dangerous situation that may occur due to changing lane.In addition, if monitoring vehicle 200 is autonomous vehicle, then control unit can be with
Keep autonomous vehicle safe by referring to the information about driving environment and the information for whether being located at blind spot about monitored vehicle
Ground changes lane.
Hereinbefore, the processing of the monitored vehicle in confirmation blind spot is illustrated about unilateral lane, however, it is possible to logical
The blind spot crossed in the lane to the other side of monitoring vehicle carries out identical step to execute the quilt in each blind spot of two sides
Monitor the confirmation of vehicle.
Meanwhile blind spot monitoring device may include CNN, CNN can be by using being pre-adjusted extremely by learning device
What a few deconvolution parameter, at least one sorting parameter and at least one frame regression parameter returned to execute convolution, classification and frame
Operation.
Here, learning device can be identical as blind spot monitoring device, but is not limited to blind spot monitoring device.Learning device and blind spot prison
If visual organ can be divided into equipment for drying to execute its function, however, for convenience, this specification description is embodied as single device
Learning device.Learning device according to the present invention may include convolutional layer, region suggest in network and FC layers etc. at least one
Part.It should be evident that similar condition can be applied to blind spot monitoring device.
It is as follows about schematically illustrating of how learning of learning device or blind spot monitoring device:
Firstly, executing the operation of convolution on training image to obtain the Feature Mapping for training, for trained spy
Sign mapping is transported to region and suggests in network, and the suggestion for training corresponding with the object being located in training image
Frame is acquired.Then, each with the Suggestion box for training by being obtained to the region on training image using pond operation
The corresponding feature vector for training of person.Later, trained feature vector will be used for and be transported at least one FC layers, and
The classification score for training with each corresponding each class of the Suggestion box for training is obtained, and obtains and uses
Information is returned in the frame for training of the corresponding each class of each of trained Suggestion box.Then, by that will classify
Number and the intended ground true value of classification are compared to obtain Classification Loss value, return by comparing frame recurrence information and frame pre-
Ground truth is determined to obtain recurrence penalty values.Then, pass through Classification Loss value acquired in backpropagation and acquired recurrence
Each of penalty values adjust deconvolution parameter, sorting parameter and frame regression parameter.
The present invention has the effect of accurately detecting the monitored vehicle being located in blind spot by using CNN.
The present invention has another effect: accurately detection is located at the monitored vehicle in blind spot without regard to monitoring vehicle
Driving environment.
The present invention also has another effect: carrying out the previous of exhaustive search with the Suggestion box to each scale and aspect ratio
Object detection network RPN on the contrary, by using minimal number of Suggestion box execute oriented mission detection come with reduction
Calculated load detection is located at the monitored vehicle in blind spot, without based on vehicle scale and the aspect ratio limited by blind spot region
Prior knowledge exhaustive search is carried out to the Suggestion box of each scale and aspect ratio.
The present invention also has another effect: the monitored vehicle in blind spot is located at by less calculating real-time detection, and
Without considering the variation of background state and visual noise.
Embodiment present invention as described above can be filled by can record the various computers of computer-readable medium
It sets and is implemented in the form of executable program commands.Computer-readable medium can include program command, number either individually or in combination
According to file and data structure.The program command that medium is recorded can be the component specially designed for the present invention, or can be right
It is available in the technical staff of computer software fields.Computer-readable medium include such as magnetic medium of hard disk, floppy disk and tape,
It the magnet-optical medium of the optical medium of such as CD-ROM and DVD, such as CD and such as ROM, RAM and is specially designed for depositing
The hardware device of the flash memory of storage and executive program order.Program command not only includes the machine language code generated by compiler,
It further include the high-level code performed by computer that can be used by interpreter etc..Above-mentioned hardware device can be used not just as using
In the software module for executing movement of the invention, they can equally be used in the opposite case.
As described above, passed through such as detailed component, limited embodiment and attached drawing specific item to the present invention
It is illustrated.The present invention is more fully understood just for the sake of help in they.However, it will be understood by those skilled in the art that
It is, it, can be from specification in the case where the spirit and scope of the present invention limited in without departing substantially from following claims
Carry out various changes and change.
Therefore, thought of the invention should not be limited to embodiment described, following Patent right requirement and including
It is equal with Patent right requirement or all the elements of equivalent variation belongs to thought category of the invention.
Claims (22)
1. a kind of method monitored by using blind spot of the blind spot monitoring device to monitoring vehicle, comprising the following steps:
(a) under conditions of the blind spot monitoring device completes or support another device to complete following processing: (i) obtain with apart from institute
It states monitoring vehicle unit distance and the one or more being located in the blind spot of the monitoring vehicle is relevant for adopting with reference to vehicle
The rear video image of sample, (ii) creation are described opposite with reference to vehicle in each of the rear video image for sampling
The reference block is set as m Suggestion box by the reference block answered, and (iii), and the m Suggestion box is used as the monitoring vehicle
Blind spot in candidate region, wherein the candidate region has the probability for detecting at least one monitored vehicle, then if from
Operation monitoring vehicle in obtain for test rear video image, then the blind spot monitoring device from it is described for test backsight
Obtain or support another device to obtain at least one Feature Mapping for test in frequency image;
(b) blind spot monitoring device (i) is suggested by the way that pond operation is applied to m in the Feature Mapping for being used to test
Frame obtains or supports another device obtain the feature vector for being used to test corresponding with each of the m Suggestion box
Each, (ii) input or another device input feature vector for test is supported to be fully connected at least one
Layer, (iii) obtain or support another device to obtain each class corresponding with each of the m Suggestion box and be used to surveying
The classification score of examination, and (iv) obtain or support another device acquisition corresponding with each of the m Suggestion box
The recurrence information for test of each class;
(c) blind spot monitoring device executes or support that another device executes following processing: (i) is by referring to described for test
Classification score selects j Suggestion box in the m Suggestion box, and (ii) is by using each of with the j Suggestion box
The every of acquisition of information bounding box corresponding with each of the j Suggestion box is returned for test described in corresponding
One, (iii) confirm each of described bounding box whether with the matching of its corresponding Suggestion box in the j Suggestion box
Degree is equal to or more than first threshold, and (iv) determines any whether monitored vehicle is located in the j Suggestion box
In person, thereby determine that whether the monitored vehicle is located in the blind spot of the monitoring vehicle.
2. according to the method described in claim 1, further comprising the steps of:
If at least one for (d) being less than second threshold with the matching degree of the corresponding Suggestion box in the j Suggestion box is special
Determine bounding box to be confirmed as finding, then the blind spot monitoring device sets or support another device to set the specific border frame conduct
Be added to include the new element of the group of the candidate region at least one dynamic Suggestion box, to obtain update group.
3. according to the method described in claim 2, wherein, in step (d), if from the rear video image for test
T frame obtain the specific border frame, then the blind spot monitoring device setting or another device is supported to set the dynamic Suggestion box
It is to include in the update group in the t+1 frame of the rear video image for test with the m Suggestion box.
4. according to the method described in claim 3, wherein, under the following conditions: (i) described monitored vehicle is confirmed as position
In from t+1 frame to t+ in the dynamic Suggestion box of (k-1) frame;And (ii) described monitored vehicle is confirmed as not being located at t+
In the dynamic Suggestion box on k frame, then the blind spot monitoring device is determining or another device is supported to determine the monitored vehicle
Whether it is located in the m Suggestion box on t+k+1 frame.
5. according to the method described in claim 1, further comprising the steps of:
(e) the blind spot controller is single by sending control for the information about the monitored vehicle being located in blind spot
Member, come support described control unit with prevent the monitoring vehicle be confirmed as being located at towards the monitored vehicle it is blind
Change lane on the direction of point.
6. according to the method described in claim 1, wherein, in step (c), the blind spot monitoring device: (i) calculates or supports another
One device calculates the first overlapping region, and in first overlapping region, each of the bounding box corresponding j are built
View frame overlaps each other;And (ii) determines or supports that another device is determining Chong Die with described first equal to or more than third threshold value
Some corresponding Suggestion box of specific first overlapping region in region be include the monitored vehicle.
7. according to the method described in claim 1, wherein, in step (c), the blind spot monitoring device: (i) calculates or supports another
One device calculates the second overlapping region, opposite with each of the j Suggestion box respectively in second overlapping region
The bounding box answered overlaps each other;(ii) determine or support another device to determine and be confirmed to be equal to or more than the 4th threshold value
At least one of second overlapping region at least one corresponding specific border frame of specific second overlapping region be packet
Include single identical monitored vehicle;And (iii) determines or supports another device to determine corresponding Suggestion box overlapping
Selected bounding box in the maximum specific border frame in region be include the monitored vehicle.
8. according to the method described in claim 1, wherein, in step (c), the blind spot monitoring device: (i) calculates or supports another
One device calculates the second overlapping region, opposite with each of the j Suggestion box respectively in second overlapping region
The bounding box answered overlaps each other;(ii) determine or support another device to determine and be confirmed to be less than described in the 5th threshold value
The corresponding specific border frame in specific second overlapping region in second overlapping region be include each monitored vehicle;With
And it includes each monitored vehicle that (iii), which is determined or another device is supported to determine that the specific border frame is,.
9. described blind under conditions of filling is set as zero in step (a) according to the method described in claim 1, wherein
Point monitor executes or supports another device to execute filter by sliding with predetermined step width to the backsight for test
Frequency image application convolution algorithm is used for the Feature Mapping of test using volume described in the rear video image acquisition to using
Product operation.
10. according to the method described in claim 1, wherein, in step (a), by with the rear video for test
The corresponding subject image application convolution algorithm of blind spot in image, the blind spot monitoring device obtain or support another device to obtain
The Feature Mapping for test.
11. according to the method described in claim 1, wherein, the blind spot monitoring device by learning device by using being adjusted
At least one deconvolution parameter, at least one sorting parameter and at least one frame regression parameter come execute convolution, classification and frame return
Return, and
Wherein, the learning device executes or supports another device to execute following processing:
(i) by obtaining the Feature Mapping for training to training image application convolution algorithm, by the spy for being used for training
Sign mapping is input to region and suggests network, and obtains corresponding with the object being located in the training image for trained
Suggestion box, (ii) apply pond operation by the region to the training image corresponding with the Suggestion box for training
Come obtain with for training each of Suggestion box it is corresponding for training feature vector each, be used for described
Trained feature vector is input to each for being fully connected layer, and obtains opposite with each of Suggestion box for training
Classify score and each class corresponding with each of Suggestion box for training for training for each class answered
The recurrence information for training, (iii) is obtained by comparing the intended ground true value of the classification score and the classification
Classification Loss value obtains recurrence penalty values by comparing the intended ground true value that the recurrence information and the frame return, with
And the convolution is learnt by each of Classification Loss value acquired in backpropagation and acquired recurrence penalty values
Parameter, the sorting parameter and the frame regression parameter.
12. a kind of blind spot monitoring device that the blind spot for monitoring vehicle is monitored, comprising:
Communication unit, under the following conditions: obtaining and monitor vehicle unit distance and positioned at the blind of the monitoring vehicle with apart from described
One or more in point is with reference to the relevant rear video image for sampling shot from the monitoring vehicle of vehicle;Creation with
It is described with reference to the corresponding reference block of vehicle in each of the rear video image for sampling;And by the reference
Frame is set as m Suggestion box, and the m Suggestion box is used as the candidate region in the blind spot of the monitoring vehicle, wherein the time
Favored area has the probability for detecting at least one monitored vehicle, then the communication unit is obtained from the monitoring vehicle of operation
Or another device is supported to obtain the rear video image for test;And
Processor, for executing or supporting another device to execute following processing: (i) is from the rear video image for test
Obtain at least one for test Feature Mapping;(ii) by a with the m in the Feature Mapping for test
The each in the corresponding region of Suggestion box is corresponding with each of the m Suggestion box to obtain using pond operation
The each of feature vector for test corresponding with each of the m Suggestion box described will be used to testing
The each of feature vector is input at least one and is fully connected layer, each of (ii-1) acquisition and the m Suggestion box
The classification score for test of corresponding each class, (ii-2) obtain corresponding with each of the m Suggestion box
Each class for test recurrence information;And (iii) is a in the m by referring to the classification score for test
J Suggestion box is selected in Suggestion box, described is used to testing by using corresponding with each of the j Suggestion box
Acquisition of information bounding box corresponding with each of the j Suggestion box is returned, confirms each of described bounding box
Whether with the matching degree of its corresponding Suggestion box in the j Suggestion box be equal to or more than first threshold, and determine described in
Whether monitored vehicle is located in any one of described j Suggestion box, thereby determines that whether the monitored vehicle is located at institute
In the blind spot for stating monitoring vehicle.
13. blind spot monitoring device according to claim 12 further includes following processing:
(iv) if being less than at least one spy of second threshold with the matching degree of the corresponding Suggestion box in the j Suggestion box
Determine bounding box to be confirmed as being found, then by the specific border frame be set as being added to include the candidate region group
At least one dynamic Suggestion box of new element, to obtain update group.
14. blind spot monitoring device according to claim 13, wherein if from the t of the rear video image for test
Frame obtains the specific border frame, then the processor sets or support another device to set the dynamic Suggestion box and the m
A Suggestion box is to include in the update group in the t+1 frame of the rear video image for test.
15. blind spot monitoring device according to claim 14, wherein under the following conditions: the monitored vehicle is true
It is set to and is located at from t+1 frame to t+ in the dynamic Suggestion box of (k-1) frame;And the monitored vehicle is confirmed as not being located at
In the dynamic Suggestion box on t+k frame, then the processor is determining or another device is supported to determine that the monitored vehicle is
In the no m Suggestion box positioned on t+k+1 frame.
16. blind spot monitoring device according to claim 12 further includes following processing:
(v) by sending control unit for the information about the monitored vehicle being located in blind spot, to support the control
Unit processed is to prevent the monitoring vehicle from changing on the direction for being confirmed as be located at blind spot towards the monitored vehicle
Lane.
17. blind spot monitoring device according to claim 12, wherein in processing (iii), the processor calculates the first weight
Folded region, in first overlapping region, j corresponding Suggestion box of each of the bounding box overlaps each other, and
And the processor will be opposite with specific first overlapping region in first overlapping region for being equal to or more than third threshold value
Some Suggestion box answered is determined as including the monitored vehicle.
18. blind spot monitoring device according to claim 12, wherein in processing (iii), the processor: calculate second
Overlapping region, in second overlapping region, the bounding box corresponding with each of the j Suggestion box respectively
It overlaps each other;By be confirmed to be at least one of described second overlapping region specific second equal to or more than the 4th threshold value
At least one corresponding specific border frame of overlapping region is determined as including single identical monitored vehicle;It will be corresponding
Selected bounding box in the maximum specific border frame in region of Suggestion box overlapping is determined as including the monitored vehicle
?.
19. blind spot monitoring device according to claim 12, wherein in processing (iii), the processor: calculate second
Overlapping region, in second overlapping region, the bounding box corresponding with each of the j Suggestion box respectively
It overlaps each other;By be confirmed to be it is corresponding less than specific second overlapping region in second overlapping region of the 5th threshold value
Specific border frame be determined as including each monitored vehicle;The specific border frame is determined as to include each monitored vehicle
?.
20. blind spot monitoring device according to claim 12, wherein in processing (i), be set as zero condition in filling
Under, the processor is by the filter that is slided with predetermined step width to the rear video image application convolution algorithm for test
Or to the Feature Mapping application convolution algorithm used described in the rear video image acquisition for test.
21. blind spot monitoring device according to claim 12, wherein in processing (i), the processor by with it is described
For test described in being obtained for the corresponding subject image application convolution algorithm of blind spot in the rear video image of test
Feature Mapping.
22. blind spot monitoring device according to claim 12, wherein the processor has been learnt by using learning device
At least one deconvolution parameter, at least one sorting parameter and at least one frame regression parameter execute or support another device to hold
Row convolution, classification and frame return, and
Wherein, the learning device executes or supports another device to execute following processing:
(a) by obtaining the Feature Mapping for training to training image application convolution algorithm, by the spy for being used for training
Sign mapping is input to region and suggests network, and obtains corresponding with the object being located in the training image for trained
Suggestion box (b) applies pond operation by the region to the training image corresponding with the Suggestion box for training
Obtain each of the feature vector for training corresponding with each of Suggestion box for training, it will be described
It is input to each for being fully connected layer for trained feature vector, and obtains each with the Suggestion box for training
Corresponding each class for the classification score of training and corresponding every with each of Suggestion box for training
The recurrence information for training of a class, (c) is obtained by comparing the classification score and the intended ground true value of the classification
Classification Loss value is taken, obtains recurrence penalty values by comparing the intended ground true value that the recurrence information and the frame return,
And the volume is learnt by each of Classification Loss value acquired in backpropagation and acquired recurrence penalty values
Product parameter, the sorting parameter and the frame regression parameter.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US15/724,544 US9934440B1 (en) | 2017-10-04 | 2017-10-04 | Method for monitoring blind spot of monitoring vehicle and blind spot monitor using the same |
US15/724,544 | 2017-10-04 |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109606263A true CN109606263A (en) | 2019-04-12 |
CN109606263B CN109606263B (en) | 2022-03-11 |
Family
ID=61711642
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811167644.4A Active CN109606263B (en) | 2017-10-04 | 2018-10-08 | Method for monitoring blind spot of monitoring vehicle and blind spot monitor using the same |
Country Status (5)
Country | Link |
---|---|
US (1) | US9934440B1 (en) |
EP (1) | EP3466764B1 (en) |
JP (1) | JP6668435B2 (en) |
KR (1) | KR102098131B1 (en) |
CN (1) | CN109606263B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111493443A (en) * | 2019-01-31 | 2020-08-07 | 斯特拉德视觉公司 | Method and device for monitoring blind area of two-wheel vehicle by using intelligent helmet |
US20220335727A1 (en) * | 2021-03-05 | 2022-10-20 | Tianiin Soterea Automotive Technology Limited Company | Target determination method and apparatus, electronic device, and computer-readable storage medium |
Families Citing this family (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116968732A (en) | 2018-03-20 | 2023-10-31 | 御眼视觉技术有限公司 | System and method for navigating a host vehicle |
CN108764292B (en) * | 2018-04-27 | 2022-03-18 | 北京大学 | Deep learning image target mapping and positioning method based on weak supervision information |
US11099579B2 (en) * | 2018-05-31 | 2021-08-24 | Nissan North America, Inc. | System for determining the number of remote vehicles following a host vehicle |
CN109166094B (en) * | 2018-07-11 | 2022-03-25 | 华南理工大学 | Insulator fault positioning and identifying method based on deep learning |
EP3787947A2 (en) | 2018-08-14 | 2021-03-10 | Mobileye Vision Technologies Ltd. | Systems and methods for navigating with safe distances |
CN112534473A (en) * | 2018-08-22 | 2021-03-19 | 日立汽车***株式会社 | Image processing apparatus and image processing method |
US10311338B1 (en) * | 2018-09-15 | 2019-06-04 | StradVision, Inc. | Learning method, learning device for detecting lanes on the basis of CNN and testing method, testing device using the same |
US10474930B1 (en) * | 2018-10-05 | 2019-11-12 | StradVision, Inc. | Learning method and testing method for monitoring blind spot of vehicle, and learning device and testing device using the same |
US10984262B2 (en) * | 2018-10-08 | 2021-04-20 | StradVision, Inc. | Learning method and testing method for monitoring blind spot of vehicle, and learning device and testing device using the same |
JP6928917B2 (en) * | 2018-11-06 | 2021-09-01 | パナソニックIpマネジメント株式会社 | Mobile management system, mobile, management device, control method, and program |
US11010907B1 (en) | 2018-11-27 | 2021-05-18 | Zoox, Inc. | Bounding box selection |
US10936902B1 (en) * | 2018-11-27 | 2021-03-02 | Zoox, Inc. | Training bounding box selection |
US10509987B1 (en) * | 2019-01-22 | 2019-12-17 | StradVision, Inc. | Learning method and learning device for object detector based on reconfigurable network for optimizing customers' requirements such as key performance index using target object estimating network and target object merging network, and testing method and testing device using the same |
US10325179B1 (en) * | 2019-01-23 | 2019-06-18 | StradVision, Inc. | Learning method and learning device for pooling ROI by using masking parameters to be used for mobile devices or compact networks via hardware optimization, and testing method and testing device using the same |
US11087175B2 (en) * | 2019-01-30 | 2021-08-10 | StradVision, Inc. | Learning method and learning device of recurrent neural network for autonomous driving safety check for changing driving mode between autonomous driving mode and manual driving mode, and testing method and testing device using them |
US10796206B2 (en) * | 2019-01-31 | 2020-10-06 | StradVision, Inc. | Method for integrating driving images acquired from vehicles performing cooperative driving and driving image integrating device using same |
JP7085504B2 (en) * | 2019-02-27 | 2022-06-16 | 本田技研工業株式会社 | Vehicle control system |
DE102019115092A1 (en) * | 2019-06-05 | 2020-12-10 | Bayerische Motoren Werke Aktiengesellschaft | Determining an object recognition rate of an artificial neural network for object recognition for an automated motor vehicle |
CN111340092B (en) * | 2020-02-21 | 2023-09-22 | 浙江大华技术股份有限公司 | Target association processing method and device |
US11537819B1 (en) * | 2020-04-30 | 2022-12-27 | Zoox, Inc. | Learned state covariances |
US10970598B1 (en) * | 2020-05-13 | 2021-04-06 | StradVision, Inc. | Learning method and learning device for training an object detection network by using attention maps and testing method and testing device using the same |
US11113574B1 (en) * | 2021-03-26 | 2021-09-07 | StradVision, Inc. | Methods for performing self-supervised learning of deep-learning based detection network by using deep Q-network and devices using the same |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050146607A1 (en) * | 2004-01-06 | 2005-07-07 | Romeo Linn | Object Approaching Detection Anti Blind e-Mirrors System |
CN101396989A (en) * | 2007-09-26 | 2009-04-01 | 日产自动车株式会社 | Vehicle periphery monitoring apparatus and image displaying method |
US20110242318A1 (en) * | 2010-03-31 | 2011-10-06 | Hon Hai Precision Industry Co., Ltd. | System and method for monitoring blind spots of vehicles |
US20130135089A1 (en) * | 2011-11-28 | 2013-05-30 | Robert Bosch Gmbh | Camera-assisted blind spot detection |
CN103237685A (en) * | 2010-12-30 | 2013-08-07 | 明智汽车公司 | Apparatus and method for displaying a blind spot |
US20140003708A1 (en) * | 2012-06-28 | 2014-01-02 | International Business Machines Corporation | Object retrieval in video data using complementary detectors |
KR20150011280A (en) * | 2013-07-22 | 2015-01-30 | 진종욱 | Driving one's view support device |
CN104369698A (en) * | 2013-08-13 | 2015-02-25 | 现代摩比斯株式会社 | Image displaying method and apparatus therefor |
CN104918825A (en) * | 2012-08-27 | 2015-09-16 | 快图有限公司 | Rearview imaging systems for vehicle |
US20170124415A1 (en) * | 2015-11-04 | 2017-05-04 | Nec Laboratories America, Inc. | Subcategory-aware convolutional neural networks for object detection |
CN106945605A (en) * | 2016-01-06 | 2017-07-14 | 北京京东尚科信息技术有限公司 | Vehicle blind zone monitoring system and control method |
CN107180220A (en) * | 2016-03-11 | 2017-09-19 | 松下电器(美国)知识产权公司 | Risk prediction method |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20150018037A (en) * | 2013-08-08 | 2015-02-23 | 주식회사 케이티 | System for monitoring and method for monitoring using the same |
US20170124409A1 (en) * | 2015-11-04 | 2017-05-04 | Nec Laboratories America, Inc. | Cascaded neural network with scale dependent pooling for object detection |
KR102592076B1 (en) * | 2015-12-14 | 2023-10-19 | 삼성전자주식회사 | Appartus and method for Object detection based on Deep leaning, apparatus for Learning thereof |
KR20170086899A (en) * | 2016-01-19 | 2017-07-27 | 연세대학교 산학협력단 | Method and Apparatus for Cyclist Detection using Multi Camera |
KR101949358B1 (en) * | 2016-04-04 | 2019-02-18 | 엘지전자 주식회사 | Apparatus for providing around view and Vehicle including the same |
US9947226B2 (en) * | 2016-04-12 | 2018-04-17 | Denso International America, Inc. | Methods and systems for blind spot monitoring with dynamic detection range |
-
2017
- 2017-10-04 US US15/724,544 patent/US9934440B1/en active Active
-
2018
- 2018-09-05 EP EP18192825.0A patent/EP3466764B1/en active Active
- 2018-10-02 KR KR1020180117319A patent/KR102098131B1/en active IP Right Grant
- 2018-10-03 JP JP2018188344A patent/JP6668435B2/en active Active
- 2018-10-08 CN CN201811167644.4A patent/CN109606263B/en active Active
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20050146607A1 (en) * | 2004-01-06 | 2005-07-07 | Romeo Linn | Object Approaching Detection Anti Blind e-Mirrors System |
CN101396989A (en) * | 2007-09-26 | 2009-04-01 | 日产自动车株式会社 | Vehicle periphery monitoring apparatus and image displaying method |
US20110242318A1 (en) * | 2010-03-31 | 2011-10-06 | Hon Hai Precision Industry Co., Ltd. | System and method for monitoring blind spots of vehicles |
CN103237685A (en) * | 2010-12-30 | 2013-08-07 | 明智汽车公司 | Apparatus and method for displaying a blind spot |
US20130135089A1 (en) * | 2011-11-28 | 2013-05-30 | Robert Bosch Gmbh | Camera-assisted blind spot detection |
US20140003708A1 (en) * | 2012-06-28 | 2014-01-02 | International Business Machines Corporation | Object retrieval in video data using complementary detectors |
CN104918825A (en) * | 2012-08-27 | 2015-09-16 | 快图有限公司 | Rearview imaging systems for vehicle |
KR20150011280A (en) * | 2013-07-22 | 2015-01-30 | 진종욱 | Driving one's view support device |
CN104369698A (en) * | 2013-08-13 | 2015-02-25 | 现代摩比斯株式会社 | Image displaying method and apparatus therefor |
US20170124415A1 (en) * | 2015-11-04 | 2017-05-04 | Nec Laboratories America, Inc. | Subcategory-aware convolutional neural networks for object detection |
CN106945605A (en) * | 2016-01-06 | 2017-07-14 | 北京京东尚科信息技术有限公司 | Vehicle blind zone monitoring system and control method |
CN107180220A (en) * | 2016-03-11 | 2017-09-19 | 松下电器(美国)知识产权公司 | Risk prediction method |
Non-Patent Citations (1)
Title |
---|
RAKESH NATTOJI RAJARAM: "RefineNet: Refining Object Detectors for Autonomous Driving", 《IEEE TRANSACTIONS ON INTELLIGENT VEHICLES,IEEE,VOL.1,NO.4》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111493443A (en) * | 2019-01-31 | 2020-08-07 | 斯特拉德视觉公司 | Method and device for monitoring blind area of two-wheel vehicle by using intelligent helmet |
CN111493443B (en) * | 2019-01-31 | 2022-09-20 | 斯特拉德视觉公司 | Method and device for monitoring blind area of two-wheel vehicle by using intelligent helmet |
US20220335727A1 (en) * | 2021-03-05 | 2022-10-20 | Tianiin Soterea Automotive Technology Limited Company | Target determination method and apparatus, electronic device, and computer-readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
KR20190039457A (en) | 2019-04-12 |
KR102098131B1 (en) | 2020-04-07 |
JP2019067405A (en) | 2019-04-25 |
US9934440B1 (en) | 2018-04-03 |
EP3466764A1 (en) | 2019-04-10 |
EP3466764B1 (en) | 2021-09-22 |
CN109606263B (en) | 2022-03-11 |
JP6668435B2 (en) | 2020-03-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109606263A (en) | To the method and the blind spot monitoring device for using this method for monitoring that the blind spot of vehicle is monitored | |
CN109624850A (en) | Monitor the method for the blind spot of vehicle and the blind spot monitoring device using this method | |
Ryan et al. | Real-time face & eye tracking and blink detection using event cameras | |
US20220114739A1 (en) | Real-time visual object tracking for unmanned aerial vehicles (uavs) | |
Venkataramanan et al. | Attention guided anomaly localization in images | |
JP7110359B2 (en) | Action Recognition Method Using Video Tube | |
US10402985B2 (en) | Collision prediction | |
CN110998594B (en) | Method and system for detecting motion | |
US20190180076A1 (en) | System and methodologies for occupant monitoring utilizing digital neuromorphic (nm) data and fovea tracking | |
US10300851B1 (en) | Method for warning vehicle of risk of lane change and alarm device using the same | |
US10706505B2 (en) | Method and system for generating a range image using sparse depth data | |
JP6650657B2 (en) | Method and system for tracking moving objects in video using fingerprints | |
EP3588372A1 (en) | Controlling an autonomous vehicle based on passenger behavior | |
US11921212B2 (en) | Long range lidar-based speed estimation | |
GB2501542A (en) | Abnormal behaviour detection in video or image surveillance data | |
CN113989702A (en) | Target identification method and device | |
Tang et al. | Hybrid blob and particle filter tracking approach for robust object tracking | |
Teoh et al. | A reliability point and kalman filter-based vehicle tracking technique | |
US10318798B2 (en) | Device and method for detecting non-visible content in a non-contact manner | |
CN109195106A (en) | Localization method and device in train | |
Liu et al. | Spatial segmentation for processing videos for farming automation | |
Marban et al. | Estimating position & velocity in 3d space from monocular video sequences using a deep neural network | |
US11496674B2 (en) | Camera placement guidance | |
Foresti et al. | Vehicle detection and tracking for traffic monitoring | |
Bansod et al. | Crowd Analysis System for Images of CCTV Camera |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |